Playground

Record a good sample for cloning

How to record a clean sample for an Instant or Professional voice clone, covering where to record, how to speak, how long to record and which files work.

A clone copies what it hears. A clean sample gives a clean voice. A noisy or mixed sample gives a noisy or mixed voice. This page is the short list of what to do before and while you record, for an Instant clone and for a Professional voice.

Before you record

  • Use your own voice only. One speaker, no one talking in the background.
  • Pick a quiet room. Close windows and doors. Turn off fans and music.
  • Avoid echo. A room with a rug, curtains or a sofa sounds better than an empty hall or a bathroom.
  • Pick the Dialect you will use the voice in. Speak in that dialect.
  • Check the browser may use your microphone.

While you record

  • Keep one steady distance from the microphone for the whole recording.
  • Speak at a natural pace, as you would in a normal talk. Do not whisper or shout.
  • Read the whole line the page shows. Do not stop in the middle.
  • If you make a mistake or a loud noise happens, press Remove and record again. A short clean take beats a long noisy one.

Instant

Press Record and read the line under Read this out loud. Keep going for 10 to 30 seconds. The page counts for you:

  • Keep reading: N more seconds until you reach 10 seconds.
  • Good. You can stop now, or keep going up to 30 seconds from 10 seconds on.

A recording stops by itself at 30 seconds. A clip over 30 seconds is refused, not trimmed. The step is on Clone your voice.

For a professional voice

  • Upload 10 minutes to 3 hours. Training needs at least 20 minutes of clear speech after cleanup, so aim for 30 minutes or more.
  • Use varied speech: different sentences, questions, numbers and a few longer passages. Do not read one sentence again and again.
  • Keep the same voice, the same microphone and the same room in every file.
  • Use short files. Up to 20 files, 2 GB in total. A recording in the browser stops at 30 minutes, so record in parts.
  • Leave out long silences, music and other speakers. They do not count as clear speech.

The steps are on Train a professional voice.

Files that work

Mode Files
Instant One file of any audio the browser can open. The page converts it. WAV, MP3 or M4A are safe choices
Professional Up to 20 files, 2 GB in total: WAV, FLAC, MP3, M4A, AAC, OGG, OPUS or WEBM

If the page cannot read an Instant file it says This file could not be read. Try a WAV, MP3 or M4A file.